CN104899601A - Identification method of handwritten Uyghur words - Google Patents
Identification method of handwritten Uyghur words Download PDFInfo
- Publication number
- CN104899601A CN104899601A CN201510288179.XA CN201510288179A CN104899601A CN 104899601 A CN104899601 A CN 104899601A CN 201510288179 A CN201510288179 A CN 201510288179A CN 104899601 A CN104899601 A CN 104899601A
- Authority
- CN
- China
- Prior art keywords
- stroke
- handwritten
- strokes
- words
- uygur
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 40
- 238000012549 training Methods 0.000 claims description 9
- 238000012937 correction Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000012952 Resampling Methods 0.000 claims description 5
- 238000005520 cutting process Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 abstract 2
- 238000005070 sampling Methods 0.000 description 9
- 230000011218 segmentation Effects 0.000 description 8
- 238000000605 extraction Methods 0.000 description 6
- 239000011159 matrix material Substances 0.000 description 4
- 230000009467 reduction Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000010365 information processing Effects 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 239000000853 adhesive Substances 0.000 description 1
- 230000001070 adhesive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Character Discrimination (AREA)
Abstract
The invention relates to an identification method of handwritten Uyghur words. The identification method comprises the following steps: firstly, preprocessing a handwritten Uyghur word; then, independently extracting a stroke structure characteristic and a gradient characteristic of the Uyghur word, wherein the stroke structure characteristic is directly extracted on a time coordinate sequence, the gradient characteristic is extracted from a two-dimensional image which is mapped by the time coordinate sequence of the preprocessed handwritten Uyghur word; carrying out serial fusion on the stroke structure characteristic and the gradient characteristic; and finally, carrying out classification identification by an Euclidean distance classifier to obtain an identification result. Two Uyghur word characteristics are extracted and are subjected to serial fusion, and an identification rate is improved than a way that single characteristic is extracted. The algorithm has the advantages of being good in algorithm performance and high in instantaneity, reliability and identification rate, is mainly applied to a mobile terminal to realize the identification of the handwritten Uyghur words, provides a new method for the informative processing of the Uyghur and opens up a new application way.
Description
Technical Field
The invention belongs to the technical field of character recognition in pattern recognition, mainly relates to the field of handwriting recognition, and particularly relates to a handwritten Uyghur word recognition method which is used for realizing the handwriting input of Uyghur words on a mobile terminal.
Background
Uyghur is an important minority language in China, is one of main characters used by Uyghur nationalities in Xinjiang, belongs to West Hungar of the Jueyang language family in the Altai language, and is an adhesive pinyin character. According to the national standard GB12050-89, modern uygur is composed of 32 letters, including 8 vowels and 24 consonants. The Uyghur characters are different in the position of the beginning, middle and end of the word due to the fact that the characters are written independently, and each character has 2-8 writing forms such as a front-connection type, a rear-connection type, a double-connection type and a single-vertical type, and evolves into 124 characters. In addition, there are two complex characters, a postamble and a diacritic, for a total of 128 Uyghur single characters. Uyghur words are composed of Uyghur single characters, and a Uyghur word is composed of 3-4 single characters in a small number and more than ten single characters in a large number. Uyghur words are written in a handwriting mode from right to left and from top to bottom, and are written along a horizontal line, namely a base line.
The study of Uyghur word processing and recognition is beneficial to the development of cultural exchange, information exchange and scientific economy in the minority nationality area of Xinjiang. In the market, the printed character recognition system for Uygur language has been widely used in various fields of Uygur language information processing, but research on recognition of handwritten words of Uygur language is still in the research stage. Compared with single-character recognition, word recognition has the following advantages: (1) more natural coherent handwriting input: when people write characters, most of the characters appear in the brain and sea by taking words as units, if the characters in one word are written one by one and recognized one by one, the continuity of thinking is easily influenced, so that the input of the words is more natural and humanized compared with the one-by-one input of single characters; (2) faster handwriting input and recognition: the handwriting recognition is directly carried out on the words, one word can be input without interruption, the overall recognition of the words is carried out, and the man-machine interaction is better embodied.
At present, the recognition of the Uyghur words is divided into a recognition method based on segmentation and a recognition method based on whole words.
The recognition method based on segmentation is characterized in that a word is regarded as a whole consisting of a plurality of single characters, the word is segmented into a series of candidate characters, then the individual characters are recognized by analyzing the shape characteristics of the characters, and then the whole word is recognized. The method has the advantages of good adaptability, no need of a large number of word training samples and storage spaces, more limitation in practical application, higher requirement on the accuracy of word segmentation, easy influence of individual difference of handwritten word writers, complex system and more strict requirement on each link in the system. Obviously, the research of the recognition method based on segmentation focuses on the character segmentation of words, the invention patent with the Chinese patent number of 201010013727.5 proposes a character segmentation method of offline Uygur language words from Lijing, Luchaoyang and the like of the university of Western electronic technology, and the invention patent proposes that segmentation is guided by multi-feature combined application, so that complex handwritten Uygur language words become clear and complete single-character images, and the clear and complete single-character images are sent to a character recognition module to realize the robust recognition of the whole Uygur language words.
The recognition method based on the whole words has the principle that starting from the whole features of the words, a global feature vector is extracted, then according to a certain matching algorithm, the global feature vector is used for matching candidate words in a known dictionary, and the candidate word with the closest distance is the final recognition result. The method has the advantages that the recognition system is simple and high in recognition speed, the problems encountered by word segmentation are avoided, and the method conforms to the habit of human reading. However, the overall feature extraction of words is difficult.
Due to the characteristic of adhesion of the Uyghur words and the randomness during writing, the extraction of the global feature vectors of the words is difficult. How to effectively combine the specific writing rule of the Uyghur words with the feature extraction algorithm and fuse different features is an urgent problem to be solved for handwriting Uyghur word recognition.
Disclosure of Invention
The invention aims to overcome the problem that the existing global feature vector for representing Uyghur words is difficult to extract, and provides a handwritten Uyghur word recognition method with high recognition rate.
The technical scheme adopted by the invention for solving the technical problems is as follows: a handwritten Uyghur word recognition method is characterized by comprising the following processing procedures:
step 1, preprocessing collected handwritten Uygur words;
step 2, mapping the Uygur language word image preprocessed in the step 1 to a feature space from an object space to obtain stroke structure features of the Uygur language word image;
step 3, mapping the time coordinate sequence of the preprocessed Uygur language words into a two-dimensional image to obtain gradient characteristics of the Uygur language word image;
step 4, fusing the stroke structure characteristics obtained in the step 2 and the gradient characteristics obtained in the step 3 to obtain a characteristic vector of the Vickers word image;
and 5, classifying and identifying the feature vectors of the Vickers words obtained in the step 4 in the feature vector library by using an Euclidean distance classifier according to the feature vector library obtained in advance by the training samples to obtain a classification and identification result.
As an improvement, the preprocessing process of the handwritten Uyghur words comprises the following steps:
(1-1) cutting the handwritten Uygur word image, and removing an area which does not contain character track points in the handwritten Uygur word image, so as to leave an area containing the character track points;
(1-2) normalizing the cut Uygur words in the step (1-1), and normalizing the handwritten Uygur word images with different sizes into images with the same size;
(1-3) performing smooth filtering on the handwritten Uygur word image subjected to the normalization processing in the step (1-2) to remove jitter noise in the handwritten Uygur word image;
(1-4) performing inclination correction on the handwritten Uygur word image processed in the step (1-3);
(1-5) resampling and interpolating trace points of the handwritten dimension word image after the inclination correction in the step (1-4), and removing the phenomenon that pixel points in the original handwritten dimension word image are compact and pixel points after normalization processing are sparse, so that the distances between the pixel points in the processed handwritten dimension word image and the pixel points in the original handwritten dimension word image are as consistent as possible.
In another improvement, the processing procedure of step 2 is as follows:
(2-1) according to the characteristics of Uygur words and the font analysis of Uygur words, referring to FIG. 3, the Uygur words are divided into 3 types of strokes: the stroke writing device comprises a main stroke, a point stroke and an additional stroke, wherein the stroke written along a base line is called the main stroke, the point stroke is a point above or below the base line, and a diacritic above the base line is the additional stroke;
(2-2) finding out main strokes in the Uygur words preprocessed in the step 1:
(2-2-1) setting a main body stroke number threshold value, wherein strokes of which the stroke points exceed the main body stroke number threshold value are main body strokes;
(2-2-2) extracting the first stroke: the first stroke after being filtered by the threshold value of (2-2-1) is the first stroke, and the first stroke is classified as the main stroke;
(2-2-3) extracting common main body strokes: directly judging the strokes of the remaining main body segments with different X coordinates from the initial strokes into common main body strokes, and classifying the common main body strokes into main body strokes;
(2-2-4) returning strokes with the beginning points tending to be closed to main strokes;
(2-3) finding out point strokes and additional strokes in the Uygur words preprocessed in the step 1: setting a point stroke number threshold, classifying strokes with the stroke number less than the point stroke number threshold as a point stroke number, classifying strokes with the stroke number more than or equal to the point stroke number threshold and less than or equal to the main stroke number threshold as an additional stroke person
(2-4) extracting direction line element characteristics of the main body strokes;
(2-5) extracting the rotation direction code characteristics of the additional strokes;
(2-6) extracting various point number characteristics of the point strokes;
the stroke structure characteristics of the Uygur language word comprise: (2-4) extracted direction line element characteristics of the main stroke, (2-5) extracted rotation direction code characteristics of the additional stroke, and (2-6) extracted point number characteristics of each type of point stroke.
The specific method for extracting the rotation direction code features of the additional strokes in the step (2-5) is as follows: connecting all coordinates of the additional strokes according to the writing sequence to obtain a curve, and extracting a rotation direction code according to the trend of the curve: the variation trends of the horizontal and vertical coordinates of the handwriting area are divided into four states: (1) Δ x is greater than or equal to 0, and Δ y is greater than or equal to 0; (2) Δ x is less than or equal to 0, and Δ y is more than or equal to 0; (3) Δ x is less than or equal to 0 and Δ y is less than or equal to 0; (4) Δ x ≧ 0, Δ y ≦ 0, and when the stroke trend is clockwise, the rotation direction code is characterized by (1) → (2) → (3) → (4) → (1); when the stroke trend is counterclockwise, the rotation direction code features are (1) → (4) → (3) → (2) → (1); when the stroke is a straight line, the rotation direction code features (1) → (1) → (1) → (1) → (1) → (1).
The specific method for extracting the number characteristics of various points of the point strokes in the step (2-6) is as follows: the number of points defining a point stroke is characterized by 7 bits: the first three bits of the characteristic represent the number of one point on the base line, the number of two points on the base line and the number of three points on the base line; the four digits after the feature respectively represent the number of one point under the baseline, the number of two horizontal points under the baseline, the number of two vertical points under the baseline and the number of three points under the baseline, and the number of various point forms is counted respectively to form the number feature of various points.
And improving, wherein the step 3 adopts a Sobel algorithm to extract the gradient characteristics of the Vickers words.
In step 4, let Z be the feature vector of the merged wiener word image, and X (X)1,x2,…,xn) Is the structural characteristic of the stroke obtained in step 2, Y (Y)1,y2,…,yn) If the gradient feature obtained in step 3 is, Z ═ α X | | | | β Y; i.e. Z ═ α x1,αx2,…,αxn,βy1,βy2,…,βyn) (ii) a Alpha is the weight coefficient of the stroke structure characteristic, beta is the weight coefficient of the gradient characteristic, which respectively represents the contribution degree of the stroke structure characteristic and the gradient characteristic to the new characteristic, and alpha is 10, and beta is 0.1.
Compared with the prior art, the invention has the advantages that: the method has the advantages of low complexity, simple realization and better robustness; the stroke structural characteristics of the invention are suitable for cursive characters such as handwritten Uygur words, can well depict the topological shape and structure of the word strokes, and has relatively small characteristic dimension and simple distance calculation; the invention effectively utilizes the advantages of different feature vectors and improves the average recognition rate of the recognition of the handwritten Uyghur words.
Drawings
FIG. 1 is an overall flow chart of the word recognition method of the present invention.
FIG. 2 is a flowchart of the present invention for the preprocessing of Uygur words.
FIG. 3 is a diagram of Uyghur word stroke model according to the present invention.
Fig. 4 is a Sobel operator template used in the present invention.
FIG. 5 is a gradient vector decomposition diagram according to the present invention.
FIG. 6 is a diagram illustrating a weight matrix sampling dimension reduction process according to the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying examples.
The embodiment provides a handwritten Uyghur word recognition method, which comprises the following steps as shown in figure 1:
step 1, preprocessing the collected handwritten Uygur words, wherein the preprocessing process is shown in figure 2 and specifically comprises the following steps:
(1-1) cutting the handwritten Uygur word image, and removing an area which does not contain character track points in the handwritten Uygur word image, so as to leave an area containing the character track points;
(1-2) normalizing the cut Uygur words in the step (1-1), and normalizing the handwritten Uygur word images with different sizes into images with the same size, wherein the size of the normalized handwritten Uygur word image is 256 multiplied by 256;
(1-3) performing smooth filtering on the handwritten Vickers word image subjected to the normalization processing in the step (1-2) to remove jitter noise in the handwritten Vickers word image; during specific implementation, a multipoint weighted average method can be adopted for smooth filtering, and the method proposes to consider the current point and the front and rear 2 points;
(1-4) performing tilt correction on the handwritten Uygur word image processed in the step (1-3) to obtain a handwritten Uygur word image; the inclination correction can adopt a Hough transformation method;
(1-5) resampling and interpolating trace points of the handwritten dimension word image after the inclination correction in the step (1-4), and removing the phenomenon that pixel points in the original handwritten dimension word image are compact and pixel points after normalization processing are sparse, so that the distances between the pixel points in the processed handwritten dimension word image and the pixel points in the original handwritten dimension word image are as consistent as possible; in this step, resampling every other 3 points can be considered;
step 2, mapping the Uygur words preprocessed in the step 1 to a feature space from an object space to obtain stroke structure features of the Uygur words, wherein the extraction process of the stroke structure features of the Uygur words comprises the following steps:
(2-1) according to the characteristics of Uygur words and the font analysis of Uygur words, referring to FIG. 3, the Uygur words are divided into 3 types of strokes: the stroke writing device comprises a main stroke, a point stroke and an additional stroke, wherein the stroke written along a base line is called the main stroke, the point stroke is a point above or below the base line, and a diacritic above the base line is the additional stroke;
(2-2) finding out the main strokes in the Uygur words preprocessed in the step 1, and the specific steps are as follows:
(2-2-1) setting a main body stroke number threshold value, wherein strokes of which the stroke points exceed the main body stroke number threshold value are main body strokes;
(2-2-2) extracting the first stroke: the first stroke after being filtered by the threshold value of (2-2-1) is the first stroke, and the first stroke is classified as the main stroke;
(2-2-3) extracting common main body strokes: directly judging the strokes of the remaining main body segments with different X coordinates from the initial strokes into common main body strokes, and classifying the common main body strokes into main body strokes;
(2-2-4) returning strokes with the beginning points tending to be closed to main strokes;
(2-3) finding out point strokes and additional strokes in the Uygur words preprocessed in the step 1: setting a point stroke number threshold, classifying strokes of which the stroke number is less than the point stroke number threshold as point stroke numbers, and classifying strokes of which the stroke number is greater than or equal to the point stroke number threshold and less than or equal to the main stroke number threshold as additional strokes; wherein the dot stroke number threshold is generally taken to be 1/10 of the normalized word width;
(2-4) extracting direction line element characteristics of the main body strokes; the direction line element characteristic is a characteristic extraction algorithm which is generally applied to handwritten character recognition, and obtains a more ideal effect in the recognition of Chinese characters. In the recognition of the handwritten Chinese words, the fact that the handwritten Chinese words have a large amount of direction information is found, and an algorithm adopting the direction characteristics is suitable; the method for extracting the direction line element characteristics of the main body strokes adopts a conventional method in the prior art, and the embodiment proposes to adopt 8 direction line element characteristics; extracting the direction line element characteristics in eight directions, and specifically explaining the following steps:
(2-4-1) calculating the direction characteristic of each sampling coordinate point in the standard direction according to the relation of the adjacent coordinate points:
for each sampling point P in the main body stroke point columnj＝(xj,yj) Its direction vectorIs defined as:
to the sampling point PiThe direction vector of (A) is vertically projected to be mapped to two standard directions of nearest neighbor, and the projection components are respectively recorded asWherein,has a large characteristic valueSmall is represented asIs expressed as a characteristic value of
If P isj＝(xj,yj) Is the intermediate point of the first image,can be calculated by the formula (4-2):
in the above formula, dx＝|xj+1-xj-1|，dy＝|yj+1-yj-1|，
If P isj＝(xj,yj) Not at intermediate points, but only for dx,dySlightly modifying the calculation of (1);
Pjthe two standard directions to which the point is projected are determined according to the specific values of the x and y coordinates, and the formula is as follows:
therefore, the direction characteristic of each stroke sampling point can be obtained; pjDirectional character of points
(2-4-2) dividing the grid, and counting the direction line element characteristics of each pixel point in the grid:
dividing the handwritten word image of the Uygur language into 8 multiplied by 8 elastic grids, and then counting the direction characteristics of all pixel points in each grid to obtain the grid direction line element characteristics;
assuming that the pixel points in a certain grid are N, N8-dimensional direction features f can be obtained1,f2,…,fN(ii) a Then, the N eigenvectors obtained are accumulated on each component according to the following formula, and the network is obtainedGrid direction line element feature vector F ═ a1,a2,a3,a4,a5,a6,a7,a8)。
(2-4-3) reducing dimension to obtain the direction line element characteristics
The grid direction line element feature is a feature vector of 512 dimensions, which is 8 × 8 × 8, and the dimension is high in the subsequent processing, so that the dimension reduction processing is required. The dimension reduction processing of sampling by using a weight matrix is adopted in the text. The weight matrix is shown in the formula (4-6).
Before sampling, a row and a column of 8 × 8 grids need to be expanded to form 9 × 9 grids, then dimension reduction processing is carried out, and the dimension of the direction line element characteristic of the main stroke of the Chinese word is 4 × 4 × 8, namely 128 dimensions;
(2-5) extracting the rotation direction code of the additional stroke, wherein the specific method comprises the following steps:
according to the observation of Uyghur words, the rotation direction of the additional part of the Uyghur words is compared regularly, in a stroke, the rotation direction is generally not more than 3, namely the linear direction, the clockwise direction and the anticlockwise direction, all coordinates of the additional strokes are connected according to the writing sequence to obtain a curve, and the rotation direction code is extracted according to the trend of the curve: the variation trends of the horizontal and vertical coordinates of the handwriting area are divided into four states: (1) Δ x is greater than or equal to 0, and Δ y is greater than or equal to 0; (2) Δ x is less than or equal to 0, and Δ y is more than or equal to 0; (3) Δ x is less than or equal to 0 and Δ y is less than or equal to 0; (4) Δ x is not less than 0 and Δ y is not more than 0; when the stroke direction is clockwise, the rotation direction code features are (1) → (2) → (3) → (4) → (1); when the stroke trend is counterclockwise, the rotation direction code features are (1) → (4) → (3) → (2) → (1); when the stroke is a straight line, the rotation direction code features are (1) → (1) → (1) → (1) → (1) → (1);
(2-6) extracting the number characteristics of various points of the point strokes, wherein the specific method comprises the following steps:
the number of points defining a point stroke is characterized by 7 bits: the first three bits of the characteristic represent the number of one point on the base line, the number of two points on the base line and the number of three points on the base line; the four digits after the feature respectively represent the number of one point under the baseline, the number of two horizontal points under the baseline, the number of two vertical points under the baseline and the number of three points under the baseline, and the number of various point forms is counted respectively to form the number feature of various points. The method comprises the following specific steps:
a) after the strokes are decomposed, the relation between the point strokes and the base line is determined according to the comparison of the maximum value and the minimum value of the point strokes in the y direction with the base line, and the maximum value of the point strokes in the y direction is set as ymaxMinimum value of yminThe base line position is ybaselineIf y ismax<ybaselineThen the point stroke is above the baseline, if ymin>ybaselineIf the point stroke is located below the baseline;
b) taking the x value of the middle point of each point stroke coordinate sequence, comparing the x values of two adjacent point strokes, and when the x values are smaller than a certain threshold value, considering the adjacent point strokes as the same point category;
c) counting the number of each category point to form various point number characteristics;
the stroke structure characteristics of the Uygur language word comprise: (2-4) extracted direction line element characteristics of the main stroke, (2-5) extracted rotation direction code characteristics of the additional stroke, and (2-6) extracted point number characteristics of each type of point stroke.
Step 3, mapping the time coordinate sequence of the preprocessed Uygur language words into a two-dimensional image to obtain gradient characteristics of the Uygur language words; in this embodiment, the Sobel algorithm is used to extract the gradient feature of the wiener word image, and the extraction of the gradient feature of the image by using the Sobel algorithm is a conventional technique, and for easy understanding, a detailed process of extracting the gradient feature of the image by using the Sobel algorithm is described as follows:
3-1), dividing the elastic grid: dividing the Uygur word image into 8 multiplied by 8 elastic grids according to a stroke density function, wherein the area grid with dense stroke density is dense, and the area grid with non-dense stroke density is sparse;
3-2), extracting a gradient direction vector of each pixel; referring to the Sobel operator in FIG. 4, the horizontal gradient value g is extracted from each pixel pointxAnd a vertical gradient value gyAnd these two gradient values are considered as the horizontal and vertical components of a gradient vector g, i.e. g ═ gx,gy]T；
3-3) decomposing the gradient direction vector of each pixel, referring to eight standard directions of FIG. 5, wherein the interval of each direction is pi/4, the gradient vector g is decomposed into two standard directions closest to the vector according to the parallelogram rule, and the characteristic value of the decomposed gradient direction is g1,g2(ii) a Only two directions have projection values, and the other directions are 0, so that a vector formed by the values in the 8 directions is the gradient characteristic of each pixel;
3-4), extracting gradient features: counting gradient vectors of each pixel in each grid of the 8 × 8 elastic grid, so as to obtain a gradient feature with dimensions of 8 × 8 × 8 being 512 dimensions;
3-5), reducing the dimensionality: since the extracted feature vectors need to be subjected to alternate point weighted sampling to complete the dimension reduction processing, the 8-score is used8, expanding the grid to 9 x 9, supplementing one row and one column to the grid respectively, assigning the characteristics of the last row and one column of the original grid to the newly added rows and columns, and using a sampling weight matrix Weighting and summing the feature vectors of the 9 × 9 grid, changing the feature vectors into a 4 × 4 grid, and obtaining feature vectors of the 4 × 4 grid, as shown in fig. 6;
the gradient feature of the Vietnamese word image obtained through the steps is 4 multiplied by 8 which is 128-dimensional;
step 4, fusing the stroke structure characteristics obtained in the step 2 and the gradient characteristics obtained in the step 3 to obtain a characteristic vector of the Vickers word image;
let Z be the fused feature vector, X (X)1,x2,…,xn) Is the structural characteristic of the stroke obtained in step 2, Y (Y)1,y2,…,yn) Is the gradient feature obtained in step 3, Z ═ α X | | | | β Y, that is, Z ═ α X | | | β Y1,αx2,…,αxn,βy1,βy2,…,βyn) (ii) a Alpha is a weight coefficient of the stroke structure characteristic, beta is a weight coefficient of the gradient characteristic, the weight coefficients respectively represent the contribution degrees of the stroke structure characteristic and the gradient characteristic to the new characteristic, alpha is 10, and beta is 0.1, and a feature vector of the Vickers word is obtained;
and 5, classifying and identifying the feature vectors of the Vickers words obtained in the step 4 in the feature vector library by using an Euclidean distance classifier according to the feature vector library obtained in advance by the training samples to obtain a classification and identification result.
In order to verify the feasibility of the method provided by the invention, firstly, 3 sets of 500-class Uygur word libraries are divided, 2 sets of the libraries are taken as training sample libraries, the remaining 1 set is taken as an identification sample library, and the 3 sets of 500-class handwritten Uygur word libraries are collected by unconstrained handwriting of Uygur people on the basis of a mobile terminal, namely a mobile phone platform; training by using 2 sets of training sample libraries to obtain a 500-class handwritten Uygur word training feature library; and (3) calculating the characteristic distance between the recognition word and each word template (500 types) by referring to the handwriting Uygur word training characteristic library and adopting an Euclidean distance classifier, and outputting a recognition result by using a minimum distance criterion.
The effect of the handwritten Uyghur word recognition method can be further illustrated by the following experimental tests:
the invention uses a word bank which is acquired by a mobile terminal, namely a mobile phone platform and is written by wiygur people without constraint to carry out experiments of identifying the Uygur words on a personal computer (Intel core i5 processor 2450M, memory 4GB and dominant frequency 2.5 GHZ). Three algorithms are adopted in the experiment, wherein the first algorithm adopts stroke structure characteristic vectors and stroke structure characteristic vector references (Wanfang. research and implementation of online handwriting Uygur character recognition technology. Xinjiang university Master academic paper (2007): 45-49.); the second algorithm uses gradient feature vectors, gradient feature reference (G.Srikantan, S.W.Lam, S.N.Srihari.Gradient-based linkage encoder for a gradient registration [ J ]. Pattern Recognition,1997,29(7): 1147-; the third algorithm adopts the handwritten Uyghur word recognition method provided by the invention. All three algorithms adopt Euclidean distance classifiers. The average recognition rate of the three algorithms for all the recognition samples is: the average recognition rate of the algorithm one is 75.2%, the average recognition rate of the algorithm two is 62.6%, and the average recognition rate of the algorithm three is 80.9%. Compared with the average recognition rate of the first algorithm, the second algorithm and the third algorithm, the method provided by the invention has higher average recognition rate, and can effectively inhibit the problem that the recognition rate is reduced due to the 'reverse stroke phenomenon'.
In summary, the handwritten Uyghur word recognition method of the invention first preprocesses the handwritten Uyghur word by cutting, normalization, smoothing, inclination correction and interpolation resampling, then extracts the online stroke structure feature and the offline gradient feature of the Uyghur word, then serially fuses the two features, and finally performs classification and recognition by adopting the Euclidean distance classifier to obtain a recognition result. The method has the advantages of good algorithm performance, strong real-time performance, high reliability, high recognition rate and the like, is mainly applied to the mobile terminal to realize handwritten Uighur recognition, provides a new method for information processing of Uighurs, and opens up a new application way.
Claims (7)
1. A handwritten Uyghur word recognition method is characterized by comprising the following processing procedures:
step 1, preprocessing collected handwritten Uygur words;
step 2, mapping the Uygur language word image preprocessed in the step 1 to a feature space from an object space to obtain stroke structure features of the Uygur language word image;
step 3, mapping the time coordinate sequence of the preprocessed Uygur language words into a two-dimensional image to obtain gradient characteristics of the Uygur language word image;
step 4, fusing the stroke structure characteristics obtained in the step 2 and the gradient characteristics obtained in the step 3 to obtain a characteristic vector of the Vickers word image;
and 5, classifying and identifying the feature vectors of the Vickers words obtained in the step 4 in the feature vector library by using an Euclidean distance classifier according to the feature vector library obtained in advance by the training samples to obtain a classification and identification result.
2. The method of recognizing handwritten Uyghur words as claimed in claim 1, wherein: the preprocessing process of the handwritten Uyghur words comprises the following steps:
(1-1) cutting the handwritten Uygur word image, and removing an area which does not contain character track points in the handwritten Uygur word image, so as to leave an area containing the character track points;
(1-2) normalizing the cut Uygur words in the step (1-1), and normalizing the handwritten Uygur word images with different sizes into images with the same size;
(1-3) performing smooth filtering on the handwritten Vickers word image subjected to the normalization processing in the step (1-2) to remove jitter noise in the handwritten Vickers word image;
(1-4) performing inclination correction on the handwritten Uygur word image processed in the step (1-3);
(1-5) resampling and interpolating trace points of the handwritten dimension word image after the inclination correction in the step (1-4), and removing the phenomenon that pixel points in the original handwritten dimension word image are compact and pixel points after normalization processing are sparse, so that the distances between the pixel points in the processed handwritten dimension word image and the pixel points in the original handwritten dimension word image are as consistent as possible.
3. The method of recognizing handwritten Uyghur words as claimed in claim 1, wherein: the processing procedure of the step 2 is as follows:
(2-1) according to the characteristics of the Uygur words and the font analysis of the Uygur words, the Uygur words are divided into 3 types of strokes: the stroke writing device comprises a main stroke, a point stroke and an additional stroke, wherein the stroke written along a base line is called the main stroke, the point stroke is a point above or below the base line, and a diacritic above the base line is the additional stroke;
(2-2) finding out main strokes in the Uygur words preprocessed in the step 1:
(2-2-1) setting a main body stroke number threshold value, wherein strokes of which the stroke points exceed the main body stroke number threshold value are main body strokes;
(2-2-2) extracting the first stroke: the first stroke after being filtered by the threshold value of (2-2-1) is the first stroke, and the first stroke is classified as the main stroke;
(2-2-3) extracting common main body strokes: directly judging the strokes of the remaining main body segments with different X coordinates from the initial strokes into common main body strokes, and classifying the common main body strokes into main body strokes;
(2-2-4) returning strokes with the beginning points tending to be closed to main strokes;
(2-3) finding out point strokes and additional strokes in the Uygur words preprocessed in the step 1: setting a point stroke number threshold, classifying strokes with the stroke number less than the point stroke number threshold as a point stroke number, classifying strokes with the stroke number more than or equal to the point stroke number threshold and less than or equal to the main stroke number threshold as an additional stroke person
(2-4) extracting direction line element characteristics of the main body strokes;
(2-5) extracting the rotation direction code characteristics of the additional strokes;
(2-6) extracting various point number characteristics of the point strokes;
the stroke structure characteristics of the Uygur language word comprise: (2-4) extracted direction line element characteristics of the main stroke, (2-5) extracted rotation direction code characteristics of the additional stroke, and (2-6) extracted point number characteristics of each type of point stroke.
4. The method of recognizing handwritten Uyghur words as claimed in claim 3, wherein: the specific method for extracting the rotation direction code features of the additional strokes in the step (2-5) is as follows: connecting all coordinates of the additional strokes according to the writing sequence to obtain a curve, and extracting a rotation direction code according to the trend of the curve: the variation trends of the horizontal and vertical coordinates of the handwriting area are divided into four states: (1) Δ x is greater than or equal to 0, and Δ y is greater than or equal to 0; (2) Δ x is less than or equal to 0, and Δ y is more than or equal to 0; (3) Δ x is less than or equal to 0 and Δ y is less than or equal to 0; (4) Δ x ≧ 0, Δ y ≦ 0, and when the stroke trend is clockwise, the rotation direction code is characterized by (1) → (2) → (3) → (4) → (1); when the stroke trend is counterclockwise, the rotation direction code features are (1) → (4) → (3) → (2) → (1); when the stroke is a straight line, the rotation direction code features (1) → (1) → (1) → (1) → (1) → (1).
5. The method of recognizing handwritten Uyghur words as claimed in claim 3, wherein: the specific method for extracting the number characteristics of various points of the point strokes in the step (2-6) is as follows: the number of points defining a point stroke is characterized by 7 bits: the first three bits of the characteristic represent the number of one point on the base line, the number of two points on the base line and the number of three points on the base line; the four digits after the feature respectively represent the number of one point under the baseline, the number of two horizontal points under the baseline, the number of two vertical points under the baseline and the number of three points under the baseline, and the number of various point forms is counted respectively to form the number feature of various points.
6. The method of recognizing handwritten Uyghur words as claimed in claim 1, wherein: and 3, extracting the gradient characteristics of the Vickers words by adopting a Sobel algorithm.
7. The method of recognizing handwritten Uyghur words as claimed in claim 1, wherein: in the step 4, Z is a feature vector of the fused Uygur word image, and X (X)1,x2,…,xn) Is the structural characteristic of the stroke obtained in step 2, Y (Y)1,y2,…,yn) If the gradient feature obtained in step 3 is, Z ═ α X | | | | β Y; i.e. Z ═ α x1,αx2,…,αxn,βy1,βy2,…,βyn) (ii) a Alpha is the weight coefficient of the stroke structure characteristic, beta is the weight coefficient of the gradient characteristic, which respectively represents the contribution degree of the stroke structure characteristic and the gradient characteristic to the new characteristic, and alpha is 10, and beta is 0.1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510288179.XA CN104899601A (en) | 2015-05-29 | 2015-05-29 | Identification method of handwritten Uyghur words |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510288179.XA CN104899601A (en) | 2015-05-29 | 2015-05-29 | Identification method of handwritten Uyghur words |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104899601A true CN104899601A (en) | 2015-09-09 |
Family
ID=54032255
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510288179.XA Pending CN104899601A (en) | 2015-05-29 | 2015-05-29 | Identification method of handwritten Uyghur words |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104899601A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295631A (en) * | 2016-07-27 | 2017-01-04 | 新疆大学 | A kind of image Uighur word recognition methods and device |
| CN106650820A (en) * | 2016-12-30 | 2017-05-10 | 山东大学 | Matching recognition method of handwritten electrical component symbols and standard electrical component symbols |
| CN106774974A (en) * | 2016-11-29 | 2017-05-31 | 网易（杭州）网络有限公司 | The method and apparatus of output information |
| CN108090489A (en) * | 2018-01-15 | 2018-05-29 | 兰州理工大学 | Offline handwriting Balakrishnan word recognition methods of the computer based according to grapheme segmentation |
| CN108573258A (en) * | 2018-04-24 | 2018-09-25 | 中国科学技术大学 | Chinese language word localization method is tieed up in a kind of quick complex background image |
| CN108664975A (en) * | 2018-04-24 | 2018-10-16 | 新疆大学 | A kind of hand-written Letter Identification Method of Uighur, system and electronic equipment |
| CN108764036A (en) * | 2018-04-24 | 2018-11-06 | 西安电子科技大学 | A kind of handwritten form Tibetan language word fourth recognition methods |
| CN108764155A (en) * | 2018-05-30 | 2018-11-06 | 新疆大学 | A kind of handwriting Uighur words cutting recognition methods |
| CN109472221A (en) * | 2018-10-25 | 2019-03-15 | 辽宁工业大学 | A kind of image text detection method based on stroke width transformation |
| CN109858480A (en) * | 2019-01-08 | 2019-06-07 | 北京全路通信信号研究设计院集团有限公司 | Digital instrument identification method |
| CN110287952A (en) * | 2019-07-01 | 2019-09-27 | 中科软科技股份有限公司 | A kind of recognition methods and system for tieing up sonagram piece character |
| CN110956175A (en) * | 2018-09-27 | 2020-04-03 | 日本电产株式会社 | Processing machine data processing device and processing machine data management system |
| CN111553336A (en) * | 2020-04-27 | 2020-08-18 | 西安电子科技大学 | Print Uyghur document image recognition system and method based on link segment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101751569A (en) * | 2010-01-15 | 2010-06-23 | 西安电子科技大学 | Character segmentation method for offline handwriting Uighur words |
| CN101866417A (en) * | 2010-06-18 | 2010-10-20 | 西安电子科技大学 | Method for identifying handwritten Uigur characters |
-
2015
- 2015-05-29 CN CN201510288179.XA patent/CN104899601A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101751569A (en) * | 2010-01-15 | 2010-06-23 | 西安电子科技大学 | Character segmentation method for offline handwriting Uighur words |
| CN101866417A (en) * | 2010-06-18 | 2010-10-20 | 西安电子科技大学 | Method for identifying handwritten Uigur characters |
Non-Patent Citations (1)
| Title |
|---|
| 瞿萌: "基于Windows Phone的联机手写维吾尔文单词识别", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295631A (en) * | 2016-07-27 | 2017-01-04 | 新疆大学 | A kind of image Uighur word recognition methods and device |
| CN106774974B (en) * | 2016-11-29 | 2019-08-13 | 网易（杭州）网络有限公司 | The method and apparatus of output information |
| CN106774974A (en) * | 2016-11-29 | 2017-05-31 | 网易（杭州）网络有限公司 | The method and apparatus of output information |
| CN106650820A (en) * | 2016-12-30 | 2017-05-10 | 山东大学 | Matching recognition method of handwritten electrical component symbols and standard electrical component symbols |
| CN106650820B (en) * | 2016-12-30 | 2020-04-24 | 山东大学 | Matching and recognizing method for handwritten electric component symbol and standard electric component symbol |
| CN108090489A (en) * | 2018-01-15 | 2018-05-29 | 兰州理工大学 | Offline handwriting Balakrishnan word recognition methods of the computer based according to grapheme segmentation |
| CN108090489B (en) * | 2018-01-15 | 2021-06-29 | 兰州理工大学 | Off-line hand-written Uygur word recognition method based on grapheme segmentation based on computer |
| CN108664975A (en) * | 2018-04-24 | 2018-10-16 | 新疆大学 | A kind of hand-written Letter Identification Method of Uighur, system and electronic equipment |
| CN108764036A (en) * | 2018-04-24 | 2018-11-06 | 西安电子科技大学 | A kind of handwritten form Tibetan language word fourth recognition methods |
| CN108573258B (en) * | 2018-04-24 | 2020-06-26 | 中国科学技术大学 | Method for quickly positioning dimension Chinese characters in complex background image |
| CN108664975B (en) * | 2018-04-24 | 2022-03-25 | 新疆大学 | Uyghur handwritten letter recognition method and system and electronic equipment |
| CN108573258A (en) * | 2018-04-24 | 2018-09-25 | 中国科学技术大学 | Chinese language word localization method is tieed up in a kind of quick complex background image |
| CN108764155B (en) * | 2018-05-30 | 2021-10-12 | 新疆大学 | Handwritten Uyghur word segmentation recognition method |
| CN108764155A (en) * | 2018-05-30 | 2018-11-06 | 新疆大学 | A kind of handwriting Uighur words cutting recognition methods |
| CN110956175A (en) * | 2018-09-27 | 2020-04-03 | 日本电产株式会社 | Processing machine data processing device and processing machine data management system |
| CN109472221A (en) * | 2018-10-25 | 2019-03-15 | 辽宁工业大学 | A kind of image text detection method based on stroke width transformation |
| CN109858480A (en) * | 2019-01-08 | 2019-06-07 | 北京全路通信信号研究设计院集团有限公司 | Digital instrument identification method |
| CN110287952A (en) * | 2019-07-01 | 2019-09-27 | 中科软科技股份有限公司 | A kind of recognition methods and system for tieing up sonagram piece character |
| CN110287952B (en) * | 2019-07-01 | 2021-07-20 | 中科软科技股份有限公司 | Method and system for recognizing characters of dimension picture |
| CN111553336A (en) * | 2020-04-27 | 2020-08-18 | 西安电子科技大学 | Print Uyghur document image recognition system and method based on link segment |
| CN111553336B (en) * | 2020-04-27 | 2023-03-24 | 西安电子科技大学 | Print Uyghur document image recognition system and method based on link segment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104899601A (en) | Identification method of handwritten Uyghur words | |
| Khalifa et al. | Off-line writer identification using an ensemble of grapheme codebook features | |
| Singh et al. | Offline script identification from multilingual indic-script documents: a state-of-the-art | |
| CN108664975B (en) | Uyghur handwritten letter recognition method and system and electronic equipment | |
| Shaw et al. | Offline Handwritten Devanagari Word Recognition: A holistic approach based on directional chain code feature and HMM | |
| Yadav et al. | Handwritten Hindi character recognition: a review | |
| Singh et al. | Online handwriting recognition systems for Indic and non-Indic scripts: a review | |
| Inunganbi et al. | Meitei Mayek handwritten dataset: compilation, segmentation, and character recognition | |
| Jangid | Devanagari isolated character recognition by using statistical features | |
| Sen et al. | Application of structural and topological features to recognize online handwritten Bangla characters | |
| James et al. | Malayalam handwritten character recognition using AlexNet based architecture | |
| Ghosh et al. | R-phoc: segmentation-free word spotting using cnn | |
| Dixit et al. | Handwritten Devanagari character recognition using wavelet based feature extraction and classification scheme | |
| Al Ghamdi | A novel approach to printed Arabic optical character recognition | |
| Bhattacharya et al. | A system for Bangla online handwritten text | |
| Gandhi et al. | An attempt to recognize handwritten Tamil character using Kohonen SOM | |
| El Moubtahij et al. | Review of feature extraction techniques for offline handwriting arabic text recognition | |
| Solanki et al. | Printed Gujarati script OCR using hopfield neural network | |
| Sundaram et al. | Performance enhancement of online handwritten Tamil symbol recognition with reevaluation techniques | |
| Jayaraman et al. | Modular approach to recognition of strokes in Telugu script | |
| Zanwar et al. | A comprehensive survey on soft computing based optical character recognition techniques | |
| Hijam et al. | Towards a complete character set Meitei Mayek handwritten character recognition | |
| Dhandra et al. | On Separation of English Numerals from Multilingual Document Images. | |
| Tan et al. | Off-line text-independent writer recognition for chinese handwriting: A review | |
| CN108416390B (en) | Hand-written script recognition methods based on two-dimensional convolution dimensionality reduction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150909 |
|
| WD01 | Invention patent application deemed withdrawn after publication |